JPH08275153A - Image compressor and image decoder - Google Patents

Abstract

PURPOSE: To obtain a compressed image and a decompressed image with high quality with a small hardware scale, simple processing, less memory capacity in which compression and decompression are conducted in real time and an edge is reserved. CONSTITUTION: The compressor and decoder is provided with a difference calculation circuit obtaining a difference between a luminance of each picture element in a block and a block representative value, a difference pattern classification circuit 4 specifying a distribution pattern of differences corresponding to each picture element of a block, and a difference representative calculation circuit 5 calculating the difference representative representing the block. In addition to a block representative value outputted from a block representative value detection circuit 2, two parameters as a difference representative value representing the block calculated by the difference and a difference distribution pattern are introduced newly as image compression data. Since the arithmetic amount is small, the scale of the hardware is small and the processing time is short.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image compression device for digitized still images or moving images and an image decompression device for decompressing an image signal compressed by the image compression device.

[0002]

2. Description of the Related Art As typical methods of lossy image compression, there are a transform coding method, a block coding method, and a vector quantization method, as shown below.

(1) Transform coding method In transform coding, an image is divided into a plurality of blocks each having (m × n) (m and n are natural numbers) pixels, and a DCT (discrete cosine) is calculated for each block. Transformation) and orthogonal transformation
After the CT coefficient is quantized, Huffman coding is performed in order of increasing frequency from the DC component.

(2) Block Coding Method In block coding, an image is divided into a plurality of (m × n) pixel blocks each having (m × n) (m and n are natural numbers) pixels. A threshold value is found for each block, and the pixels in the block are binarized with this threshold value to create a bit plane having (m × n) elements. Furthermore, the result of this binarization is 1
The representative luminance values of the groups that have become 0 and the groups that have become 0 are calculated, and the representative luminance values of these two groups and the above (m × n)
A block of (m × n) pixels is encoded with a bit plane of size.

(3) Vector Quantization Method In contrast to scalar quantization in which each sample is quantized independently, vector quantization is a method in which a plurality of samples are collectively quantized as a vector. The simplest method of applying vector quantization to image coding is to input and quantize a vector containing the luminance values of all pixels in a block of (m × n) pixels to a vector quantizer. Is the way.

Regarding the above-mentioned three methods, the fifth (Vol. 4) of the lecture "Introduction to Image Information Compression" serialized in the Journal of the Television Society, Vol. 43, No. 6 to Vol. 44, No. 6
3, No. 10 (1989)), 6th (Vol. 43, No. 11)
(1989)), and the ninth (Vol.44, No.2 (1990)).

(4) Fractal image compression method In addition to the above-mentioned three typical methods, a fractal image compression method, which is a coding method utilizing partial self-similarity of a natural image, has recently been proposed. There is a method. The partial self-similarity is a property that a pixel pattern similar to a certain pixel pattern in an image exists in the same image. Details of this encoding method are described in, for example, Image Coding Symposium (PCSJ91) 6-11 “Encoding of Images Using Fractals”.

(5) Other Systems One of the systems described in Japanese Patent Application Laid-Open No. 61-263369 is one that utilizes reduction in an image compression system that does not fall into the above-mentioned four systems. In this method, the image is divided into a plurality of blocks, the average value of the brightness values is calculated for each block to reduce the image, and each average value of the reduced image (average value image) and the corresponding original image The first stage image compression is performed by calculating the difference from the average value for each pixel in the block to form a difference image, and encoding this difference image by Huffman encoding or the like. Next, using the reduced image of the first stage as an original image, the same process as the image compression of the first stage is performed, and the image compression of the second stage is performed. The same image compression process is repeated below, and image compression is performed in multiple stages until the number of pixels of the new original image becomes 1 or until the number of pixels becomes appropriate.

[0009]

The image compression apparatus and the image decompression apparatus that employ the above-described conventional image compression method have the following problems, respectively.

(1) Transform coding method The transform coding method requires a floating-point operation and requires a large amount of calculation, which increases the scale of hardware and requires a long processing time. Further, since high frequency components are lost, there is a drawback that edges are blurred when highly compressed.

(2) Block coding method Although the circuit scale can be made relatively small, the block size is usually 4 × 4 and 16 pixels are expressed by two representative values, so that the image quality of the restored image is high. not really good.

(3) Vector Quantization Method At present, the number of vector dimensions that can be realized is 16, that is, (4 ×
It is about 4), but at this level, there is a correlation that cannot be ignored between input vectors. Therefore, it is disadvantageous in terms of coding efficiency as compared with the transform coding method capable of coding with a block size of about 8 × 8 or 16 × 16.
Further, the vector quantizer is designed by the LBG algorithm, using a vector extracted from many images as a learning sequence. However, when there is a large difference in the statistical properties of the learning sequence and the quantization target image, performance degradation occurs. It is practically impossible to design a versatile quantizer that will not cause performance degradation for most images.
(Details of the LBG algorithm are described on pages 27 to 28 of the book "Speech Recognition by Probabilistic Model" published by the Institute of Electronics, Information and Communication Engineers.).

(4) Fractal image compression method In fractal image compression, affine transformation such as reduction and rotation of a partial image must be performed, and pattern matching must be performed between the affine transformed partial image and other parts of the image. I have to. The affine transformation and pattern matching take a very long time. Also, with this scheme, the entire image must remain in memory while the image is being compressed.

(5) Multi-step image compression method described in Japanese Patent Laid-Open No. 61-263369 In this method, since the image is hierarchically reduced, there is a drawback that the processing also takes time. Also, a memory is required to store the reduced image of the first stage,
It also requires a lot of memory.

Therefore, an object of the present invention is that the processing is simple,
To provide an image compression device and an image decompression device that can be compressed / decompressed in real time and can obtain a high-quality compressed image and decompressed image that can be compressed / decompressed in real time with a small hardware scale and a small memory. is there.

[0016]

In order to achieve the above object, an image compression apparatus according to the invention of claim 1 provides a digitized original image with pixels of m rows × n columns (m and n are natural numbers). Image dividing means for dividing the plurality of blocks into a plurality of blocks, and detecting the minimum value or the maximum value of the luminance of the pixels in each block of the plurality of blocks, the minimum value or the maximum value of the luminance is the luminance of each block. Block representative value detection means for representing a block representative value, difference calculation means for obtaining a difference value between the brightness of each pixel in the block and the block representative value, and the difference corresponding to each pixel in the block From the difference value distribution pattern specifying means for specifying the value distribution pattern to any one of a plurality of predetermined distribution patterns and the difference value corresponding to each pixel in the block, Difference representative value calculating means for calculating a difference representative value representative of the block, block representative value quantizing means for quantizing the block representative value, difference representative value quantizing means for quantizing the difference representative value, and The quantized block representative value quantized by the block representative value quantizing means, the distribution pattern value representing the distribution pattern of the difference values specified by the difference value distribution pattern specifying means, and the quantization by the difference representative value quantizing means. And a coded data output means for sequentially outputting the quantized quantized difference representative value.

The image restoration apparatus according to the second aspect of the invention is
A quantized block representative value obtained by quantizing a block representative value representing each block of a plurality of blocks divided from a digitized original image, and a difference value between the brightness of each pixel in the block and the block representative value Is a distribution pattern value representing a distribution pattern distributed to each pixel in the block, and a quantized difference representative value obtained by quantizing a difference representative value representing the block calculated from the difference value,
Fixed-length code decoding means for decoding from the bit string of the compressed image data transmitted from the image compression device, and dequantization of the quantized block representative value from the fixed-length code decoding means, and the number of bits of one pixel of the original image. Block representative value dequantization means for outputting the block representative value returned to the above, and the difference representative value obtained by dequantizing the quantized difference representative value from the fixed length code decoding means and returning it to the bit number of one pixel of the original image And the block representative value from the block representative value dequantizing means as the luminance value (m rows × n
(Column) (m and n are natural numbers) enlargement means for making a block having pixels, distribution pattern values of difference values from the fixed length code decoding means, and difference representative values from the difference representative value dequantization means Is input, based on the distribution pattern value above,
Difference information adding means for detecting a pixel having a difference among (m rows × n columns) pixels included in the block, and adding the difference representative value to the luminance value of the pixel having the difference, and the difference information adding means. And a restored image output unit for outputting the block having the pixels of (m rows × n columns) restored by adding the difference information by the unit to a predetermined position for restoring the original image. There is.

The image compression apparatus according to the third aspect of the invention is
Image dividing means for dividing the digitized original image into a plurality of blocks having m rows × n columns of pixels, and detecting a minimum value or a maximum value of the luminance of the pixels in each block of the plurality of blocks. , A block representative value detecting unit that uses the minimum value or the maximum value of the luminance as a block representative value that represents the luminance of each block, and a difference that obtains a difference value between the luminance of each pixel in the block and the block representative value In the block, a calculation unit, a difference value distribution pattern specifying unit that specifies the distribution pattern of the difference values corresponding to each pixel in the block to any one of a plurality of predetermined distribution patterns Difference representative value calculating means for calculating a difference representative value representative of the block from the difference value corresponding to each pixel of the block, and a block representative value quantum for quantizing the block representative value. Means a difference representative value quantization means for quantizing said difference representative value, based on the quantized differential representative value quantized by the difference representative value quantization means,
Difference value distribution pattern correction means for correcting a distribution pattern value representing the distribution pattern of the difference values, the quantized quantized block representative value, the corrected distribution pattern value, and the quantized quantization Encoding means for converting the differential representative value into a variable length code, encoded data for sequentially outputting the quantized block representative value encoded by the encoding means, the corrected distribution pattern value, and the quantized difference representative value. And output means.

The image restoration device of the invention of claim 4 is
An encoded quantized block obtained by quantizing a block representative value representative of each block of a plurality of blocks divided from a digitized original image from a bit string of compressed image data transmitted from an image compression device The quantized block representative that represents a representative value and a distribution pattern value in which a difference value between the brightness of each pixel in the block and the block representative value is distributed to each pixel in the block The corrected distribution pattern value corrected based on the value and the encoded quantized difference representative value that is encoded after the difference representative value that is representative of the block calculated from the difference value is quantized and then the quantized Variable-length code decoding means for obtaining the coded block representative value, the modified distribution pattern value, and the quantized difference representative value, and the quantization block from the variable-length code decoding means. Table values by inverse quantization, the original image 1
The block representative value dequantizing means for outputting the block representative value returned to the bit number of the pixel and the quantized difference representative value from the variable length code decoding means are dequantized to obtain the bit number of one pixel of the original image. Difference representative value dequantization means for outputting the returned difference representative value, and enlarging means for making a block having (m rows × n columns) (m and n are natural numbers) pixels whose luminance value is the block representative value. , The modified distribution pattern value from the variable length code decoding means and the differential representative value from the differential representative value dequantization means are input, and the block has (m rows × x) based on the modified distribution pattern value. a difference information adding unit that detects a pixel having a difference among pixels of (n columns) and adds the difference representative value to the luminance value of the pixel having the difference;
And a restored image output unit for outputting a block having pixels of (m rows × n columns) restored by adding the difference information by the difference information adding unit to a predetermined position for restoring the original image. It is characterized by that.

[0020]

In the image compression apparatus according to the first aspect of the invention, the digitized original image is divided by the image dividing means into a plurality of blocks having pixels of m rows × n columns. Then, the block representative value detecting means detects the minimum value or the maximum value of the luminance of the pixels in the divided blocks, and the minimum value or the maximum value of the luminance represents the luminance of each block. It is a representative value.

Then, the difference calculating means obtains a difference value which is a difference between the brightness of each pixel in the block and the block representative value. Then, the difference value distribution pattern specifying means specifies the arrangement of the difference values corresponding to the respective pixels in the block to one of the distribution patterns classified into several types. For example, when m = 2 and n = 2,
It is detected which position in the block is two or three from the largest difference value in the block, and the detected distribution pattern is one of the prepared distribution patterns classified into several types. Select whether or not you are doing.

Further, the difference representative value calculating means calculates a difference representative value representing the block from the difference values corresponding to each pixel in the block. For example, the difference representative value is calculated by taking the average of the larger two difference values corresponding to each pixel in the block.

The block representative value quantizing means quantizes the block representative value by bit-shifting the block representative value. The difference representative value quantization means quantizes the difference representative value. For example, the difference representative value quantization means performs linear quantization by bit-shifting the difference representative value, or performs non-linear quantization by performing a lookup table conversion on the difference representative value.

Then, the encoded data output means sequentially outputs the quantized block representative value, the distribution pattern of the difference values specified by the difference pattern specifying means, and the quantized difference representative value. To do. All of these three data (the distribution pattern of the block representative value and the difference value and the difference representative value) have a fixed length.

By the way, a compression method in which the digitized original image is divided into a plurality of blocks each having (m rows × n columns) of pixels and only the representative value (block representative value) in each block is transmitted. Then, it is inevitable that the restored image becomes a blurred image due to the loss of high frequency components.

Therefore, according to the first aspect of the invention, in addition to the block representative value, a representative difference value representing a block calculated from the difference value corresponding to each pixel in the block and each pixel in the block. Two parameters, that is, the difference value distribution pattern of the difference values corresponding to are newly introduced. As a result, the high frequency component is saved in the restored image. This is the concept of the invention of claim 1.

More specifically, for example, the image dividing means may generate (m × n) (m = 2, n =) original digitized images.
In the case of dividing into a plurality of blocks having a plurality of pixels of 2), as a result of taking statistics of the difference value distribution pattern of the difference value between the brightness of each pixel in the block and the block representative value, the largest difference It was found that the third largest difference value is considerably smaller than the value and the second largest difference value.

Therefore, when the difference value distribution pattern indicating the positions of the first and second difference values from the largest difference value in the 2 × 2 pixel matrix is examined, the six patterns shown in FIG. 9 are obtained. Among them, it was found that pattern 0, pattern 1, pattern 2 and pattern 3 were statistically very large. On the other hand, it has been found that the above arrangement pattern is rarely the pattern 4 or the pattern 5.

Therefore, for example, if the number of bits assigned to the distribution pattern of difference values is to be suppressed to about 2 bits,
Patterns 4 and 5 may be ignored and only patterns 0-3 may be used. By doing so, it is possible to obtain a restored image with high image quality in which high-frequency components are saved with a small amount of data.

If the compression efficiency may be a little worse, the distribution of three difference values from the largest difference value to the third largest difference value is also considered, and the number of patterns is set to 8.
By increasing the number of patterns to 10, the image quality is further improved.

Further, in the image restoration device of the invention of claim 2, the fixed-length code decoding means sets the quantized block representative value, the distribution pattern value, and the quantized difference representative value to the quantized difference representative value. Decoding is performed from the bit string of the compressed image data transmitted from the image compression device.

Then, the block representative value dequantization means dequantizes the quantized block representative value from the fixed length code decoding means and outputs the block representative value returned to the number of bits of one pixel of the original image. To do.

The differential representative value dequantization means is
The quantized difference representative value from the fixed length code decoding means is inversely quantized and the difference representative value obtained by returning to the bit number of one pixel of the original image is output.

Then, the enlarging means uses the block representative value from the block representative value dequantizing means as the luminance value.
A block having pixels of (m rows × n columns) is created.

The difference information adding means receives the distribution pattern value of the difference values from the fixed length code decoding means and the difference representative value from the difference representative value dequantizing means and inputs the distribution pattern value. Based on the above, among the pixels of (m rows × n columns) included in the block, a pixel having a difference is detected, and the difference representative value is added to the luminance value of the pixel having the difference.

Then, the restored image output means restores the original image from a block having pixels of (m rows × n columns) restored by adding the difference information by the difference information adding means. Output to position.

The image compression apparatus according to the invention of claim 3 is
The image dividing means, the block representative value detecting means, the difference calculating means, the difference value distribution pattern specifying means, the difference representative value calculating means, the block representative value quantizing means, and the difference representative value quantizing means, It has the same function as that of the image compression apparatus of the first aspect of the invention.

Then, the difference pattern correction means included in the image compression apparatus according to the third aspect of the present invention calculates the distribution pattern value of the difference values based on the quantized difference representative value quantized by the difference representative value quantization means. Fix it.

More specifically, when the quantized quantized difference representative value is 0, a distribution pattern value indicating that there is no difference, that is, that all four pixels have the same brightness, is prepared and already determined. The difference pattern value that has been set is corrected to the distribution pattern value that indicates that "there is no difference". Then, the encoding means converts the quantized quantized block representative value, the modified distribution pattern value, and the quantized quantized difference representative value into a code such as a Huffman code. At this time, the quantized block representative value may be Huffman-encoded as it is, or the difference between the quantized block representative value and the quantized block representative value of the previous block may be Huffman-encoded (DPCM). As a coding table for Huffman coding, block representative values, distribution pattern values, and
What is created from the statistical distribution of the difference representative values may be stored in the ROM or the like.

Then, the encoded data output means sequentially outputs the quantized block representative value encoded by the encoding means, the corrected distribution pattern value, and the quantized difference representative value.

By the way, the image compression apparatus according to the first aspect of the present invention is an apparatus for performing fixed-length encoding of image data. This fixed-length encoding has a specific application, but the information to be output
The compression efficiency can be improved by Huffman coding (block representative value, distribution pattern value, difference representative value).

By including the above-mentioned distribution pattern value representing "having no difference" in a plurality of predetermined distribution patterns, the compression efficiency can be further improved. Specifically, it is assumed that the difference values of the pixels of 2 rows × 2 columns are, for example, 0, 1, 0, 1 in the raster order. The distribution pattern of the difference values in this case is specified as the distribution pattern 3 shown in FIG. 9 by the difference value distribution pattern specifying means.

However, when the differential representative value is calculated and quantized (depending on the design of the quantization table, for example, linear quantization by bit shift or taking a logarithm of two bases), the differential representative value is 0. Often becomes.
It is meaningless to add a distribution pattern value expressing a distribution pattern of difference values to such a block whose difference representative value is 0. This is because the difference representative value representing a block being 0 means that there is no difference in the luminance of the pixels in the block. Also,
When the distribution of the quantized difference representative value is examined, the difference representative value is 0 in many cases. Therefore, when the differential representative value of a block is 0, a distribution pattern value representing “having no difference” is assigned to the block, and the encoded data output means does not output the differential representative value. By doing so, the compression rate can be considerably improved.

The Huffman table used at the time of Huffman coding is obtained from the statistical data of a large number of images and is not necessarily optimum for each image, but the block representative value, distribution pattern value, difference The statistical properties such as the representative value do not change so much depending on the image.

Therefore, the deterioration of the compression rate compared with the case of using the optimum Huffman table for each image is neglected as compared with the loss of the compression rate due to the addition of the statistical information of each image to the encoded image. It is within the range.

Further, in the image decompressing device of the invention of claim 4, the decoding means encodes and quantizes the bit string (Huffman code string) of the compressed image data transmitted from the image compressing device of claim 2. The encoded quantized block representative value, the modified distribution pattern value, and the encoded quantized encoded quantized difference representative value are decoded, and the quantized block representative value and the modified distribution pattern value are decoded. And a quantized difference representative value. The Huffman tree used at the time of decoding is R that is created in advance from the distribution of block representative values of many images, distribution pattern values, and difference representative values.
It is stored in the OM or the like. Then, a block representative value dequantization means, a difference representative value dequantization means, an expansion means,
The difference information adding means and the restored image output means are the two.
It works the same as the invention of and restores the image.

[0047]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The image compression apparatus and the image decompression apparatus of the present invention will be described in detail below with reference to the illustrated embodiments.

[First Embodiment] FIG. 1 shows a schematic configuration of a first embodiment of an image compression apparatus of the present invention. This image compression device is for compressing digital image data input by an image input device such as a CCD camera not shown. The digital image data may be 8-bit data or 16-bit data, for example. Hereinafter, description will be made assuming that the digital image data has 8 bits. It should be noted that the diagonal line segments provided in the approximate center of the arrows connecting the circuits in FIG. 1 and 8, 6 and 3 described beside them indicate the number of bits of data transmitted there, respectively. 8 bits, 6 bits, and 3 bits are shown. This also applies to FIGS. 2, 3, and 5.

In FIG. 1, 1 is an image division circuit, 2 is a block representative value detection circuit, 3 is a difference calculation circuit, 4 is a difference pattern classification circuit as a difference value distribution pattern specifying means,
Reference numeral 5 is a difference representative value calculation circuit, 6 is a block representative value quantization circuit, 7 is a difference representative value quantization circuit, and 8 is an encoded data output circuit.

An input image is input to the image division circuit 1 as a digitized original image. Then, the image division circuit 1 divides the input image into a plurality of blocks each having pixels of 2 rows × 2 columns.

From this image division circuit 1, only the luminance data of the block having pixels of 2 rows × 2 columns need be sequentially input to the block representative value detection circuit 2. Therefore, the image division circuit 1 may be configured as shown in FIG. That is, the image division circuit 1 includes a line memory 1a, a 1-pixel delay circuit 1b connected in series to the line memory 1a, and a 1-pixel delay circuit 1 connected in parallel to the line memory 1a.
and c. According to this image division circuit 1, FIG.
As shown in, the luminance data at the 1st row and 1st column from the terminal a is 1
It is delayed by the clock and then output.
The luminance data of the column is output. Further, the luminance data of the second row and the first column is delayed from the terminal c by one clock and outputted, and the luminance data of the second row and the second column is outputted from the terminal d. As described above, by configuring the image division circuit 1 to have the circuit configuration shown in FIG. 6, the processing after the block representative value detection circuit 2 is made to rest for one clock, and after the processing of one line is completed, the next one line is processed. It is sufficient to realize the sequential input of the above-mentioned luminance data by taking a break. By doing so, the memory capacity of the image dividing circuit 1 is enough for one line. Therefore, according to the first embodiment, the image compression device can be realized with a small memory.

Next, the block representative value detection circuit 2 selects the brightness value a, the brightness value b, the brightness value c, and the brightness value d of the pixels of 2 rows × 2 columns included in the block output from the image division circuit 1. The minimum value of is detected as a block representative value. In addition, this first
In the embodiment, the block representative value detection circuit 2 detects the minimum value of the brightness values a, b, c, d as the block representative value, but the maximum value of the brightness values as the block representative value. You may make it detect.

On the other hand, the difference calculation circuit 3 calculates the difference between the block representative value and the luminance values a, b, c, d of the pixels of 2 rows × 2 columns, that is, the difference values Da, Db, Dc, Dd. .

Next, in the difference pattern classification circuit 4, the difference value Da between the luminance value of the pixels of 2 rows × 2 columns and the block representative value Da
From Db, Dc, Dd, the distribution pattern of the difference value of the block is obtained. The distribution pattern of this difference value is the difference value D
2 or 3 from the larger of a, Db, Dc, and Dd is 2.
It depends on the position of the pixel of row × 2 columns. Then, the value of the determined distribution pattern is determined depending on which distribution pattern of the prepared distribution patterns the determined distribution pattern matches.
In the first embodiment, the larger one of the difference values is 2
The distribution pattern is determined by the position of one difference value with respect to the pixel array of 2 rows × 2 columns. This distribution pattern is specifically shown in FIG. In FIG. 9, six distribution patterns from pattern 0 to pattern 5 are shown. In each distribution pattern, in the array of pixels of 2 rows × 2 columns, the positions of the pixels having two difference values from the one having the largest difference value are represented by the shaded rectangles. There is. As shown in FIG. 9, when the digitized original image is divided into a plurality of blocks having a plurality of (m × n) (m = 2, n = 2) pixels, As a result of the statistics of the difference value distribution pattern of the difference values between the pixel brightness and the block representative value, the third largest difference value is considerably smaller than the largest difference value and the second largest difference value. I knew that.

Therefore, when the difference value distribution pattern showing the positions occupied by the first and second difference values from the largest difference value in the 2 × 2 pixel matrix is examined, the six patterns shown in FIG. 9 are obtained. Among them, it was found that pattern 0, pattern 1, pattern 2 and pattern 3 were statistically very large. On the other hand, it has been found that the above arrangement pattern is rarely the pattern 4 or the pattern 5.

Therefore, for example, if the number of bits assigned to the distribution pattern of difference values is to be suppressed to about 2 bits,
Patterns 4 and 5 may be ignored and only patterns 0-3 may be used. By doing so, it is possible to obtain a restored image with high image quality in which high-frequency components are saved with a small amount of data.

If the compression efficiency may slightly deteriorate, the distribution of three difference values from the largest difference value to the third largest difference value is also taken into consideration, and the number of patterns is set to 8.
By increasing the number of patterns to 10, the image quality is further improved.

The difference representative value calculation circuit 5 has 2 rows ×
A difference representative value is calculated from the difference values Da, Db, Dc, and Dd corresponding to the pixels in the two columns. In this embodiment, the representative difference value is a value obtained by averaging two difference values from the larger difference value Da, Db, Dc, Dd.

Next, the difference representative value quantization circuit 7 quantizes the 8-bit difference representative value received from the difference representative value calculation circuit 5 into about 3 bits or 4 bits. This quantization may be linearly quantized by, for example, bit shift, or when nonlinear quantization such as a logarithmic function is performed, it may be realized by lookup table conversion. In the first embodiment, the quantization bit number is 3 bits, and the quantization is performed non-linearly by the look-up table conversion.

On the other hand, the block representative value quantizing circuit 6 receives the block representative value from the block representative value detecting circuit 2 and quantizes the block representative value. This quantization is performed, for example, by shifting the block representative value, which is 8-bit data, to the lower order by about 2 bits.

Next, as shown in FIG. 7, the encoded data output circuit 8 quantizes the block representative value (6 bits).
And the difference pattern number as the distribution pattern value of the difference value
A bit string formed by (3 bits) and the quantized difference representative value (3 bits) is output as one block of encoded image data.

According to the first embodiment, in addition to the block representative value output by the block representative value detection circuit 2, the difference representative representing the block calculated from the difference value corresponding to each pixel in the block. Two parameters, that is, a value and a difference value distribution pattern (difference pattern number) of difference values corresponding to each pixel in the block are newly introduced. As a result, it is possible to compress the image data with high quality while suppressing the deterioration of the image quality without losing the high frequency component.

Further, according to the first embodiment, unlike the transform coding method, there is no need to perform floating point arithmetic and the amount of arithmetic is small, so that the scale of hardware is small and the processing time is short. is there.

Further, according to the first embodiment, since the affine transformation and the pattern matching processing are not performed, the processing time is shorter than that of the fractal image compression, and the image compression can be performed in real time.

[Second Embodiment] FIG. 2 shows a schematic configuration of a second embodiment as an embodiment of the image restoration apparatus of the present invention. The image restoration apparatus of the second embodiment receives the encoded image compressed by the image compression apparatus of the first embodiment shown in FIG.
This is a device for restoring the input coded image. Figure 2
In the figure, 9 is a fixed-length code decoding circuit, 10 is a block representative value inverse quantization circuit, 11 is a difference representative value inverse quantization circuit, 1
Reference numeral 2 is an enlargement circuit, 13 is a difference information addition circuit, and 14 is a restored image output circuit.

The fixed-length code decoding circuit 9 divides the bit string of the input coded image into 6 bits, 3 bits, and 3 bits, and quantizes the block representative value, the difference pattern number as the distribution pattern value, and the quantization. The difference representative value is output.

Next, the block representative value inverse quantization circuit 10
Shifts the 6-bit quantized block representative value input from the fixed length code decoding circuit 9 to the right by 2 bits and returns it to 8 bits. That is, the block representative value inverse quantization circuit 1
0 performs processing (inverse quantization) reverse to that when the block representative value is quantized in the first embodiment.

Next, the difference representative value inverse quantization circuit 11 dequantizes the 3-bit quantized difference representative value from the decoding circuit 9 to restore it to an 8-bit difference representative value. In the first embodiment, since the quantization is performed by the look-up table conversion, the dequantization in the second embodiment performs the look-up table conversion that is the reverse of the quantization.

Next, the enlarging circuit 12 receives the block representative value returned to 8 bits by the block representative value dequantizing circuit 10, and outputs the 8-bit block representative value to pixels of 2 rows × 2 columns. Expand to a block with.

Next, the difference information adding circuit 13 detects the distribution pattern value from the decoding circuit 9 and the difference representative value inverse quantization circuit 1
The partial image formed by the pixels of 2 rows × 2 columns received from the enlargement circuit 12 by using the differential representative value from 1 (8 bits)
Add difference information to (block). That is, by the pattern obtained from the difference pattern number, 2 rows × 2
For pixels that are determined to have a difference among the pixels in the column,
The difference representative value is added to the brightness value of this pixel.

Next, the restored image output circuit 14 outputs the partial image (block) formed by the pixels of 2 rows × 2 columns restored by the above processing to a predetermined position where the original image is restored. .

The image compressed by the first embodiment can be restored by repeating the series of image processing described above until the bit string of the input coded image ends.

As described above, according to the second embodiment, the fixed length code decoding circuit 9 converts the bit string of the compressed image data transmitted from the image compression apparatus of the first embodiment into
The block representative value, the difference pattern number, and the difference representative value are decoded to form a block having 2 rows × 2 columns of pixels with the block representative value as a luminance value, and the pixel of the block is determined based on the difference pattern number. The difference representative value is added. Therefore, according to the second embodiment, the image data compressed by the first embodiment can be restored.

According to the second embodiment, since the image data is restored by using the two parameters of the difference representative value and the difference pattern number in addition to the block representative value, the small hardware scale and the small amount are used. With the amount of calculation,
It is possible to restore high-quality image data while suppressing deterioration of image quality without losing high-frequency components.

Further, according to the second embodiment, unlike the transform coding method, there is no need to perform floating point arithmetic and the amount of arithmetic is small, so that the scale of hardware is small and the processing time is short. is there.

Further, according to the second embodiment, since the affine transformation and the pattern matching processing are not performed, the processing time is shorter than that in the case where the fractal image compressed image data is restored, and the image compression can be performed in real time. it can.

[Third Embodiment] Next, FIGS. 3 and 4 show a third embodiment. The right ends x, y, z in FIG. 3 are connected to the left ends x, y, z in FIG. 4, respectively.

The third embodiment is an image compression device. In the third embodiment, for example, 8-bit digital image data input from an image input device such as a CCD camera, which is not shown, is compressed.

As shown in FIGS. 3 and 4, the third embodiment differs from the first embodiment shown in FIG. 1 only in the following three points. That is, the difference pattern correction circuit 15 connected between the difference pattern classification circuit 4 as the difference pattern identification means and the encoded data output circuit 18, the block representative value quantization circuit 6, and the difference pattern A Huffman encoder 16 is connected between the correction circuit 15, the differential representative value quantization circuit 7 and the encoded data output circuit 8, and a ROM 17 including a Huffman table is connected to the Huffman encoder 16. That is the point.

Therefore, the image division circuit 1, the block representative value detection circuit 2, the difference calculation circuit 3, the difference pattern classification circuit 4, the difference representative value calculation circuit 5, the block representative value quantization circuit 6, and the difference The representative value calculation circuit 7 has the same configuration as that of the first embodiment and has the same function as that of the first embodiment. Therefore, also in the third embodiment, the image division circuit 1 divides the input image into a plurality of blocks each having pixels of 2 rows × 2 columns.

Therefore, in the third embodiment, the difference pattern correction circuit 15, the Huffman encoder 16 and the ROM are used.
17 will be mainly described.

The difference pattern correction circuit 15 corrects the difference pattern number as the distribution pattern value specified by the difference pattern classification circuit 4 based on the quantized difference representative value quantized by the difference representative value quantization circuit 7. To do. Specifically, the difference pattern correction circuit 15 has a seventh distribution pattern different from the six difference value distribution patterns shown in FIG. 9, and the seventh distribution pattern is included in the block. The four pixels have the same luminance value. The seventh distribution pattern represents that the blocks have no difference. Therefore,
The difference pattern correction circuit 15 assigns the difference pattern number of the seventh distribution pattern to the block whose quantized difference representative value is 0.

On the other hand, the Huffman encoder 16 quantizes the block representative value from the block representative value quantization circuit 6, the difference pattern number from the difference pattern correction circuit 15, and the difference representative value quantization circuit 7. Upon receiving the quantized difference representative value, the block representative value, the distribution pattern value, and the difference representative value are Huffman-encoded, respectively. In this Huffman coding, the block representative value may be Huffman coded as it is, or DPCM
(Differential PCM) may be used. The Huffman table used for the Huffman coding is stored in the ROM 17. As shown in FIG. 4, the Huffman table is
Huffman table for block representative values for Huffman coding block representative values, Huffman table for distribution pattern values for Huffman coding distribution pattern values, and difference representative values for Huffman coding difference representative values Includes a Huffman table and.

Each Huffman table is created from the distribution of block representative values for many images, the distribution of difference pattern numbers for many images, and the distribution of difference representative values for many images. .

The Huffman encoder 16 gives the ROM 17 an address calculated from the data of each of the block representative value, the difference pattern number, the difference representative value and the offset of the Huffman table, and reads each encoded data. At this time, when the value of the difference value distribution pattern is a value representing a distribution pattern having no difference, the Huffman encoder 16 does not encode the difference representative value. In addition, the differential representative value 0 is also stored in the differential representative value Huffman table.
There is no code equivalent to.

Then, the encoded data output circuit 18 outputs the block representative value encoded by the Huffman encoder 16, the difference pattern number, and the difference representative value as a bit string shown in FIG. 8A. The bit lengths of the block representative value, the difference pattern number, and the difference representative value shown in FIG. 8A are indefinite, and the difference representative value may not be output as shown in FIG. 8B. Therefore, although not shown in FIG.
The data passed from the Huffman encoder 16 to the encoded data output circuit 18 includes not only the encoded block representative value, the encoded difference pattern number, and the encoded difference representative value, but also these respective bits. It also includes the length information.

As described above, in the third embodiment, by performing the variable length coding such as the Huffman coding, the compression rate is improved as compared with the first embodiment in which the fixed length coding is performed. The reason is that Huffman coding is a code that can shorten the code length per data on average by giving a short code to data that frequently appears and giving a long code to data that rarely appears. It is because it is a system. In addition to this, the third embodiment introduces a distribution pattern number indicating that there is no difference, and when the difference representative value is 0, does not output the difference representative value, thereby further improving the compression rate. You can Fourth Embodiment Next, FIG. 5 shows a schematic configuration of a fourth embodiment which is an embodiment of the image restoration apparatus of the present invention. The fourth embodiment is an image decoding apparatus that decodes a coded image compressed by the image compression apparatus of the third embodiment shown in FIGS. 3 and 4. The fourth embodiment is a variable length code decoding circuit 19
2 is different from the second embodiment shown in FIG. 2 only in that the ROM 30 connected to the variable code decoding circuit 19 and including the Huffman tree is provided. Therefore, this point will be mainly described.

In FIG. 5, 19 is a variable length code decoding circuit, and 30 is a ROM. Then, the block representative value dequantization circuit 10, the difference representative value dequantization circuit 11, the expansion circuit 12, the difference information addition circuit 13, and the restored image output circuit 14
Has the same configuration as that of the second embodiment and operates in the same way as the circuit of the second embodiment.

The variable length code decoding circuit 19 is provided in the ROM 3
Using the Huffman tree (Huffman tree for block representative value, Huffman tree for difference pattern, and Huffman tree for difference representative value) stored in 0, the block representative value and the distribution pattern from the bit string of the input coded image. The difference pattern number as a value and the difference representative value are decoded. As described in the third embodiment, when the difference pattern number is a value representing a pattern having no difference, the difference representative value is not included in the input bit string and is not decoded.

Next, the block representative value inverse quantization circuit 10,
The differential representative value dequantization circuit 11, the expansion circuit 12, the difference information addition circuit 13, and the restored image output circuit 14 have the same configuration as in the second embodiment of FIG. 2 and operate in the same manner as in the second embodiment.

Then, as in the second embodiment, the decoding circuit 1
Until the bit string of the encoded image input to 9 ends, the decoding circuit 19 outputs the block representative value, the distribution pattern value,
Decode the differential representative value. Then, the block representative value dequantization circuit 10 and the difference representative value dequantization circuit 11 dequantize the block representative value and the difference representative value. Then, the expansion circuit 12 expands the block representative value into pixels of 2 rows × 2 columns. Then, the difference information addition circuit 13 repeats the process of outputting the partial image restored after the difference information is added from the restored image output circuit 14 to restore the compressed image.

According to the image restoration apparatus of the fourth embodiment,
When the difference pattern number is a value representing a pattern having no difference, the difference representative value is not included in the input bit string and is not decoded. Therefore, according to the fourth embodiment, in addition to the effect of the image decompression device of the second embodiment, the image data from the image compression device of the third embodiment having higher compression efficiency than that of the first embodiment is decompressed. Thus, high quality image data can be output in real time.

[Fifth Embodiment] Next, FIG. 10 shows a system overview of a digital still camera incorporating a compression chip 32 and a decompression chip 34 as an embodiment of an image compression device and an image decompression device of the present invention. Show. As shown in FIG. 10, this digital still camera is used by connecting to a host device 35 realized by a personal computer or a portable information tool.

In FIG. 10, reference numeral 31 denotes an image input section, which includes an optical system section, a CCD drive system section, a γ correction section, an A / D conversion section and the like. The image input unit 31 outputs a digitized image. The image input unit 31 includes the compression chip 32.
Are connected, and an image memory 33 is connected to the compression chip 32. Then, in the image memory 33,
The restoration chip 34 is connected. A personal computer as a host device 35 or a portable information tool is connected to the restoration chip 34.

In this digital still camera, the digital image input to the image input unit 31 is processed by the image compression chip 32.
And the compressed image data is stored in the image memory 33.

On the other hand, when the image is read from the host device 35 side, the image data stored in the image memory 33 is
The host device 3 is restored by the image restoration chip 34.
5 is transmitted. As described above, the compression chip 32 and the decompression chip 34 that realize the image compression device and the image decompression device of the present invention have a simple circuit configuration and can compress and decompress image data in real time. Therefore, the user does not feel a time lag due to the image processing time when capturing or reading an image using this digital still camera, and is aware that the image is compressed and decompressed. Without doing so, a speedy operation feeling can be obtained.

When the first embodiment in which the image compression device of the first aspect of the invention is realized is built in the image decompression chip 34, the image data is compressed to about 1/3. To be done. Further, when the image decompression chip 34 is provided with the third embodiment in which the image compression apparatus according to the invention of claim 3 is realized, the image data is reduced to 1/3 to 1/3.
It is compressed in half (1/4 to 1/5 on average).

Therefore, according to the still camera shown in FIG. 10, the apparent memory capacity of the image memory in the camera body is apparent in comparison with the still camera having no image compression / decompression device shown in FIG. It has the effect of being used 3 to 5 times.

[0099]

As is apparent from the above description, the image compression apparatus according to the first aspect of the present invention comprises a difference calculation means for obtaining a difference value between the brightness of each pixel in the block and the block representative value.
Difference value distribution pattern specifying means for specifying the distribution pattern of the difference values corresponding to each pixel in the block to one of a plurality of predetermined distribution patterns, and to each pixel in the block. Difference representative value calculating means for calculating a difference representative value representative of the block from corresponding difference values, a quantized quantized block representative value, the distribution pattern value, and a quantized quantized difference. And a coded data output means for sequentially outputting the representative value.

Therefore, according to the image compressing apparatus of the invention of claim 1, in addition to the block representative value, a difference representative value representative of the block calculated from the difference value corresponding to each pixel in the block, Two parameters, that is, a difference value distribution pattern of difference values corresponding to each pixel in the block are newly introduced. As a result, high-frequency components are not lost and edges are preserved, so that it is possible to suppress deterioration in image quality and perform high-quality image data compression without blurring.

According to the first aspect of the present invention, unlike the transform coding method, it is not necessary to perform a floating point arithmetic operation and can be realized by only an integer arithmetic operation.
Since complicated processing such as T is not performed, the circuit configuration is simple, the hardware scale is small, the processing time is short, real-time image compression is possible, and the memory is small. Further, since the fixed length coding is used, partial restoration of the image is also possible.

The image restoration device of the invention of claim 2 is
Fixed-length code decoding means for decoding the quantized block representative value, the distribution pattern value, and the quantized difference representative value from the bit string of the compressed image data transmitted from the image compression device, and the block representative value as the luminance value. The enlarging means for creating a block having (m rows × n columns) of pixels and the distribution pattern value of the difference value and the representative difference value are input, and the block has (m rows × n columns) based on the distribution pattern value. Of the pixels having the difference, the difference information adding means for detecting the pixel having the difference and adding the difference representative value to the luminance value of the pixel having the difference, and the difference information being added by the difference information adding means and restored (m rows) ×
and a restored image output means for outputting a block having pixels of (n columns) to a predetermined position where the original image is restored.

Therefore, according to the invention of claim 2, the restoration of the image compressed by the image compressing apparatus of the invention of claim 1 can be realized only by integer arithmetic operation and by simple arithmetic operation. The circuit configuration is simple, and compressed images can be restored in real time.

The image compression apparatus according to the invention of claim 3 is
The difference calculation means for obtaining the difference value between the brightness of each pixel in the block and the block representative value, and the distribution pattern of the difference value corresponding to each pixel in the block are one of a plurality of predetermined distribution patterns. Difference value distribution pattern specifying means for specifying one, difference representative value calculating means for calculating a difference representative value representing a block from difference values corresponding to each pixel in the block, and quantized quantization Difference value distribution pattern correction means for correcting the distribution pattern value of the difference value based on the difference representative value, the quantized quantized block representative value, the corrected distribution pattern value, and the quantized quantized difference Encoding means for converting the representative value into a code,
It is provided with a coded data output means for sequentially outputting the quantized block representative value variable-length coded by the coding means, the corrected distribution pattern value, and the quantized difference representative value.

The image compression apparatus according to the third aspect of the present invention can realize the image compression only by the integer operation as in the first aspect of the invention, and does not perform complicated processing such as DCT (discrete cosine transform). Therefore, the calculation amount is small, the circuit configuration is simple,
It can be compressed in real time and requires less memory. Then, since the compression is performed with the edges preserved, a high-quality decompressed image without blur can be obtained. Furthermore, the image compression apparatus according to the third aspect of the present invention uses an encoding means for performing variable length encoding such as Huffman encoding, and does not output this when the differential representative value is 0. can do.

Therefore, according to the image compression apparatus of the third aspect of the present invention, a higher compression rate can be obtained than that of the image compression apparatus of the first aspect of the present invention, and a high compression rate can be expected especially for an image with few high frequency components.

The image restoration apparatus according to the fourth aspect of the invention is
It is for restoring the compressed and encoded image data from the image compression device of the invention of claim 3.

This image restoration device can realize image restoration only by integer arithmetic, like the image restoration device according to claim 2.
Further, since it can be realized only by a simple calculation, the circuit configuration is simple and the compressed image can be restored in real time.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an image compression apparatus that is a first embodiment of the present invention.

FIG. 2 is a schematic configuration diagram of an image restoration device that is a second embodiment of the present invention.

FIG. 3 is a schematic configuration diagram of an image compression apparatus that is a third embodiment of the present invention.

FIG. 4 is a schematic configuration diagram of an image compression apparatus that is a third embodiment of the present invention.

FIG. 5 is a schematic configuration diagram of an image restoration device that is a fourth embodiment of the present invention.

FIG. 6 is a diagram showing a configuration example of an image division circuit of the image compression apparatus of the first embodiment.

FIG. 7 is a diagram showing an output format of encoded data of the image compression apparatus of the first embodiment.

FIG. 8 is a diagram showing an output format of encoded data of the image compression apparatus of the third embodiment.

FIG. 9 is a diagram showing six difference value distribution patterns.

FIG. 10 is a diagram showing the configuration of a digital still camera provided with an image compression apparatus as a fifth embodiment of the present invention.

FIG. 11 is a diagram showing a configuration of a digital still camera that does not include an image compression device.

[Explanation of symbols]

1 ... Image division circuit, 2 ... Block representative value detection circuit, 3 ...
Difference calculation circuit, 4 ... Difference pattern classification circuit, 5 ... Difference representative value calculation circuit, 6 ... Block representative value quantization circuit, 7 ... Difference representative value quantization circuit, 8 ... Encoded data output circuit, 9 ...
Fixed-length code decoding circuit, 10 ... Block representative value inverse quantization circuit, 11 ... Difference representative value inverse quantization circuit, 12 ... Enlargement circuit,
13 ... Difference information addition circuit, 14 ... Restoration image output circuit, 1
5 ... Difference pattern correction circuit, 16 ... Huffman encoder, 1
7 ... ROM, 18 ... Encoded data output circuit, 19 ... Variable length decoding circuit, 30 ... ROM, 1a ... Line memory, 1b ...
1 pixel delay, 1c ... 1 pixel delay.

Claims (4)

[Claims] 1. The original digitized image is represented by m rows × n columns.
Image dividing means for dividing into a plurality of blocks having (m and n are natural numbers) pixels, and a minimum or maximum value of the luminance of pixels in each block of the plurality of blocks to detect the minimum value of the luminance. Alternatively, a block representative value detecting means having a maximum value as a block representative value representing the luminance of each block, a difference calculating means for obtaining a difference value between the luminance of each pixel in the block and the block representative value, and the block A difference value distribution pattern specifying unit that specifies the distribution pattern of the difference values corresponding to each pixel in the block to any one of a plurality of predetermined distribution patterns; Difference representative value calculating means for calculating a difference representative value representative of the block from the difference value, block representative value quantizing means for quantizing the block representative value, Difference representative value quantizing means for quantizing the minute representative value, quantized block representative value quantized by the block representative value quantizing means, and distribution pattern of difference values specified by the difference value distribution pattern specifying means And an encoded data output unit that sequentially outputs the distribution pattern value representing the above and the quantized difference representative value quantized by the difference representative value quantization unit. 2. A quantized block representative value obtained by quantizing a block representative value representing each block of a plurality of blocks divided from a digitized original image, the luminance of each pixel in the block, and the block representative. A quantized difference obtained by quantizing a distribution pattern value representing a distribution pattern in which a difference value with a value is distributed to each pixel in the block, and a difference representative value representing the block calculated from the difference value. The representative value and the fixed-length code decoding means for decoding from the bit string of the compressed image data transmitted from the image compression device, and the quantization block representative value from the fixed-length code decoding means are inversely quantized to obtain the original image. Block representative value dequantization means for outputting the block representative value returned to the number of bits of one pixel, and the original image by dequantizing the quantized difference representative value from the fixed length code decoding means. Difference representative value dequantization means for outputting the difference representative value returned to the number of bits of one pixel, and the block representative value from the block representative value dequantization means as brightness values (m rows × n columns) (m And n are natural numbers) enlargement means for making a block having pixels, a distribution pattern value of difference values from the fixed-length code decoding means, and a difference representative value from the difference representative value dequantization means are inputted, Difference information for detecting a pixel having a difference among (m rows × n columns) pixels included in the block based on the distribution pattern value and adding the difference representative value to the luminance value of the pixel having the difference Adding means and restored image output means for outputting a block having (m rows × n columns) pixels restored by adding the difference information by the difference information adding means to a predetermined position where the original image is restored It is characterized by having and Image recovery device. 3. The original digitized image is represented by m rows × n columns.
Image dividing means for dividing into a plurality of blocks having (m and n are natural numbers) pixels, and a minimum or maximum value of the luminance of pixels in each block of the plurality of blocks to detect the minimum value of the luminance. Alternatively, a block representative value detecting means having a maximum value as a block representative value representing the luminance of each block, a difference calculating means for obtaining a difference value between the luminance of each pixel in the block and the block representative value, and the block A difference value distribution pattern specifying unit that specifies the distribution pattern of the difference values corresponding to each pixel in the block to any one of a plurality of predetermined distribution patterns; Difference representative value calculating means for calculating a difference representative value representative of the block from the difference value, block representative value quantizing means for quantizing the block representative value, Difference representative value quantizing means for quantizing the minute representative value, and a distribution pattern value representing a distribution pattern of the difference values is corrected based on the quantized difference representative value quantized by the difference representative value quantizing means. Difference value distribution pattern correction means, coding means for converting the quantized quantized block representative value, the corrected distribution pattern value, and the quantized quantized difference representative value into a variable length code. And an encoded data output means for sequentially outputting the quantized block representative value encoded by the encoding means, the corrected distribution pattern value and the quantized difference representative value. . 4. A block representative value representing each block of a plurality of blocks divided from a digitized original image is quantized and encoded from a bit string of compressed image data transmitted from an image compression apparatus. The coded quantized block representative value and the distribution pattern value representing the distribution pattern in which the difference value between the luminance of each pixel in the block and the block representative value is distributed to each pixel in the block, A modified distribution pattern value that is modified based on the block representative value that has been converted, and a coded quantized difference representative value that is encoded by quantizing the difference representative value that represents the block calculated from the difference value. A variable length code decoding means for decoding and obtaining a quantized block representative value, the modified distribution pattern value, and a quantized difference representative value; Block representative value dequantization means for dequantizing the child block representative value and outputting the block representative value returned to the number of bits of one pixel of the original image; and the quantized difference representative value from the variable length code decoding means. Is inversely quantized to output the differential representative value that is returned to the bit number of one pixel of the original image, and the block representative value is used as the luminance value (m rows × n columns) (m And n are natural numbers) enlargement means for making a block having pixels, and a modified distribution pattern value from the variable length code decoding means,
The difference representative value from the difference representative value dequantization means is input, and a pixel having a difference is detected from (m rows × n columns) pixels included in the block based on the corrected distribution pattern value. , A block having difference information adding means for adding the difference representative value to the luminance value of the pixel having this difference, and (m rows × n columns) pixels restored by adding the difference information by the difference information adding means. And a restored image output unit that outputs the original image to a predetermined position so as to restore the original image.